Analyzing School Effects: How to Use the Same Data to Support Different Hypotheses

Abstract

In the usual school effects studies three types of variables can be distinguished: (a) school measures (i.e., characteristics of the school such as financial resources and characteristics of the instructional program); (b) input variables (i.e., the student characteristics at the time he enters the system); and (c) output variables (i.e., student characteristics after being in the system for a given period of time). In these studies the school effect is considered to be the association between the school measure and output when input is controlled. In recent years the number of statistical techniques used to control for input has multiplied to such a degree that the researcher has a difficult time choosing the best procedure for his study. For example, within the framework of regression analysis, the following methods have been recommended by one or more authors: Method 1. The partial correlation between the school variable and the output, controlling input variables. Method 2. The proportion of variance the school variable adds to prediction after all input variables have entered in a stepwise regression routine. The square root of this proportion of variance is also used and is shown in Appendix A to be the part correlation between the school variable and the output with input variables partialed out of the school variable. This variance may be computed by subtracting the squared multiple correlation of input with output from the squared multiple correlation of all variables with output.